Intelligent Partitioning Method for Free-form grid Structure Based on Generative Adversarial Networks

Abstract

The diverse and irregular nature of architectural free-form surfaces presents a significant challenge in grid partitioning. The prevalence of explicitly coding-oriented methods tailored to specific free-form surface features has led to a lack of comprehensive exploration of the broader intrinsic patterns of these surfaces, thereby limiting the overall generality of these methods. Therefore, we introduce GridGAN to uncover the inherent patterns in the data and facilitate the grid partitioning for free-form surfaces. First, we innovatively propose utilizing the curvature and height cloud maps of free-form surfaces as inputs to GridGAN to obtain the corresponding grid structures. Then, we utilize self-developed codes to facilitate the data dimension conversion and obtain the desired free-form grid structures. Lastly, the exceptional merits of our proposed method have been thoroughly substantiated through 12 test samples and 2 cases and a comprehensive evaluation that seamlessly combines qualitative evaluation (expert perception) with quantitative evaluation (FID score and geometric properties) methods.